[mlir][transform] Emit error message with emitSilenceableFailure

[srcarroll]

The previous implementation used a notifyMatchFailure to emit failure message inappropriately and then used the emitDefaultSilenceableFailure. This patch changes this to use the more appropriate emitSilenceableFailure with error message. Additionally a failure test has been added.

[llvmbot]

@llvm/pr-subscribers-mlir

@llvm/pr-subscribers-mlir-linalg

Author: None (srcarroll)

Changes

The previous implementation used a notifyMatchFailure to emit failure message inappropriately and then used the emitDefaultSilenceableFailure. This patch changes this to use the more appropriate emitSilenceableFailure with error message. Additionally a failure test has been added.

---

Full diff: https://github.com/llvm/llvm-project/pull/86146.diff

2 Files Affected:

• (modified) mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp (+4-5)
• (added) mlir/test/Dialect/Linalg/flatten-unsupported.mlir (+16)

diff --git a/mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp b/mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp
index ecf9983124821a..06acade06d771b 100644
--- a/mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp
+++ b/mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp
@@ -3269,11 +3269,10 @@ DiagnosedSilenceableFailure transform::FlattenElementwiseLinalgOp::applyToOne(
     transform::ApplyToEachResultList &results,
     transform::TransformState &state) {
   rewriter.setInsertionPoint(target);
-  if (!isElementwise(target)) {
-    failed(rewriter.notifyMatchFailure(
-        target, "only elementwise flattening is supported"));
-    return emitDefaultSilenceableFailure(target);
-  }
+  if (!isElementwise(target))
+
+    return mlir::emitSilenceableFailure(target->getLoc())
+           << "only elementwise flattening is supported";
   // If rank <= 1, do nothing
   if (target.getNumLoops() <= 1) {
     results.push_back(target);
diff --git a/mlir/test/Dialect/Linalg/flatten-unsupported.mlir b/mlir/test/Dialect/Linalg/flatten-unsupported.mlir
new file mode 100644
index 00000000000000..476733fc74f5cf
--- /dev/null
+++ b/mlir/test/Dialect/Linalg/flatten-unsupported.mlir
@@ -0,0 +1,16 @@
+// RUN: mlir-opt %s -transform-interpreter -split-input-file -verify-diagnostics
+
+func.func @non_elementwise(%arg0: memref<2x3xf32>, %arg1: memref<3x4xf32>, %arg2: memref<2x4xf32>) {
+  // expected-error @+1 {{only elementwise flattening is supported}}
+  linalg.matmul ins(%arg0, %arg1 : memref<2x3xf32>, memref<3x4xf32>) outs(%arg2: memref<2x4xf32>)
+  return
+}
+
+module attributes {transform.with_named_sequence} {
+  transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {
+    %0 = transform.structured.match interface{LinalgOp} in %arg1 : (!transform.any_op) -> !transform.any_op
+    %flattened = transform.structured.flatten_elementwise %0
+      : (!transform.any_op) -> !transform.any_op
+    transform.yield
+  }
+}

[srcarroll]

I'm still waiting on my local build to test this change

[srcarroll]

@ftynse As a general rule of thumb, should i even bother waiting for the windows buildkite to finish? often it just hangs for several hours, other times it just fails for reasons irrelevant to the PR, and rarely does it actually finish and pass (for me at least)

[srcarroll]

need to fix this one too https://github.com/llvm/llvm-project/pull/86146/files#diff-61bfc4a79549b25af3e46ff8002c9a9aea61fb537b565435ba0957e2b53b9adaR3286. just need to figure out a test case to trigger it

[srcarroll]

need to fix this one too https://github.com/llvm/llvm-project/pull/86146/files#diff-61bfc4a79549b25af3e46ff8002c9a9aea61fb537b565435ba0957e2b53b9adaR3286. just need to figure out a test case to trigger it

@ftynse this was done after your approval so i'll wait for you thumbs up on this

[ftynse]

@ftynse As a general rule of thumb, should i even bother waiting for the windows buildkite to finish? often it just hangs for several hours, other times it just fails for reasons irrelevant to the PR, and rarely does it actually finish and pass (for me at least)

Use your best judgement and be ready to revert if submitting before it passes. I usually wait when changing the build system (Windows/MSVC tends to catch missing dependencies) or ABI-impacting things, but may not wait otherwise. 10 hours is an overkill, but it did pass this time.

[ftynse]

need to fix this one too https://github.com/llvm/llvm-project/pull/86146/files#diff-61bfc4a79549b25af3e46ff8002c9a9aea61fb537b565435ba0957e2b53b9adaR3286. just need to figure out a test case to trigger it

Thanks!